Staying in sync with a network filesystem
OFS uses synchronization in two situations: to keep the cache up to date while the server connection is up and to reintegrate any local changes with the server when a lost connection is re-established.
Alternative tools, such as Coda and Intermezzo, provide software on the server side to let the server report changes to all relevant clients. OFS, of the other hand, is a client-only solution – it is the user's responsibility to trigger the cache update process manually.
When updating the cache, OFS relies on the files' modification timestamps. Then the OFS daemon compares the timestamp of the original file in the share to the copy in the cache. If the remote file has changed, OFS copies it to the local cache.
To synchronize, OFS must perform a full search of the offline directory tree. Because this process stresses the network, the hard disk, and the CPU, the developers decided to do without full automation based on polling.
When a program issues an open() system call, the update process is triggered. At that point, OFS checks to see whether the file in its offline directory is up to date; if not, it downloads the latest version from the server. OFS uses the local cache to answer read access requests, whereas write requests modify both the cache and the share (Figure 4). This approach guarantees both fast read access and up-to-date files.
When the server connection is down, OFS logs all write accesses handled by the cache. Each entry in the logfile contains both the path to the file and a parameter specifying the change. The possible values are: created, deleted, and modified.
Once the server connection is restored, OFS works its way through the logfile, performing each of the steps on the server side: creating new files, deleting existing files, or uploading the cached version to the server.
During reintegration, a file the user has modified locally might have been changed on the server, too. To avoid simply overwriting the file, OFS first checks how current the remote file is. If the modification time is later than the last synchronization time, the remote file must have been changed in the meantime. In this case, OFS launches its conflict-resolution tool.
In some cases – for example, if one user has modified the local file while another user has simply renamed the version on the server – the conflict is easily resolved. However, some conflicts are impossible to resolve automatically, such as cases in which two users modify the file in different ways while it is offline. In this situation, OFS leaves the decision to the users. A diff mechanism analyzes both versions of the file, and a preview shows the results of the proposed conflict resolution. (The OFS configuration decides which diff tool to use.) Text files are typically fairly easy to merge, with the exception of a situation in which both sides have changed the same part of the file. Because it is more difficult to display and merge differences in binary files, the only approach is to choose one of the versions.
The KFilePlugin plugin for Dolphin and Konqueror adds an OFS tab to the Properties dialog field, which is displayed when you right-click a file or directory and select Properties. By selecting the OFS tab, users can add a directory to the local cache or write changes out to the server. This GUI plugin is not a required part of OFS. Machines without a GUI can still use OFS at the command line.
For the most part, OFS interacts with applications just like any other filesystem. Its special features include a flag that specifies whether a file should be available offline if the connection is lost. Many filesystems, AFS for example, have their own tools for setting these parameters. OFS, however, removes the need for special tools by relying on extended file attributes, which the user can set and read through standard Linux file utilities.
Extended file attributes make it possible to add metadata to files (as long as the filesystem supports this). The kernel does not handle the get and set commands itself but passes them directly to the filesystem. It is thus the filesystem's responsibility to process or save the attributes. FUSE supports extended file attributes and offers callback functions for setting, deleting, reading, and listing attributes.
OFS relies on this feature to implement its own custom attributes, which serve only to exchange information between the filesystem and the shell. OFS does not save them and does not pass them on to the underlying filesystem; you might say they are virtual attributes. The current version of OFS supports two attributes in the ofs namespace:
- ofs.offline indicates whether a file is available offline. Setting this attribute triggers the update process and adds the tree to the list of directories available offline.
- ofs.available indicates whether the file is available right now, or whether the cached version is in use. In other words, it tells you whether the server connection is available. This attribute is read-only.
In both cases, the only two states are "set" (yes) and "not set" (no). setfattr -n attribute_path sets the state, and getfattr -n attribute_path reads it. To delete the attribute completely, you can use setfattr -x attribute_path.
OFS is not a mature product, although the current version is usable if you accept its limitations. The roadmap features a number of extensions. The modular design supports the introduction of more agent classes, for example, to provide more precise details on the availability of a remote filesystem.
- OFS (Offline filesystem): http://offlinefs.sourceforge.net
- FUSE: http://fuse.sourceforge.net
- Fusexx, C++--Bindings for FUSE: http://portal.itauth.com/2007/07/07/c-fuse-binding
- D-Bus: http://dbus.freedesktop.org
- Windows offline files : http://support.microsoft.com/kb/307853/
- Coda: http://www.coda.cs.cmu.edu
- Subversion: http://subversion.tigris.org
- Unison: http://www.cis.upenn.edu/~bcpierce/unison/
Buy this article as PDF
New partnership will bring more and better CS training to US schools
Criminals offer online help over Tor network
Sophisticated malware is still present on Joomla and WordPress sites around the world.
Future versions of Ubuntu's code service will support the popular Git version control system used with Linux and other open source projects.
New release marks the arrival of AMD’s unified driver strategy.
A new study by IDC charts big changes in the big hardware market.
Azure CTO says Redmond has already considered the unthinkable.
Lead developer quells rumors that the Debian version is slated for center stage.
MSBuild is now just another GitHub project as Redmond continues its path to the light.
Malware could pass data and commands between disconnected computers without leaving a trace on the network.